OpenAI Disrupts Russian and Iranian AI Disinformation Ops

OpenAI has exposed and banned ChatGPT accounts linked to two state-backed influence campaigns from Russia and Iran that used artificial intelligence to trick legitimate news organizations into publishing propaganda. Unlike typical covert operations that rely primarily on bot networks across social media platforms, these campaigns relied on fictitious identities to insert fake stories directly into traditional media ecosystems.

The Russian operation, designated Dark Clark, focused on destabilizing politics in Latin America and discrediting Ukraine. Operating through a bogus think tank managed by a fake persona, the group likely recruited local staff who were unaware of the scheme. Dark Clark generated fabricated audio files and forged documents that spurred official government denials and formal fact-checks in Ecuador and Peru. OpenAI rated Dark Clark as a category 5 threat on its 6-point Breakout Scale because local politicians actively reacted to the synthetic media. This marks the first category 5 incident OpenAI has documented in two and a half years of threat monitoring.

The Iranian campaign, dubbed Bogus Bylines, deployed seven fake journalists to publish almost 100 articles covering the conflict between the United States and Iran in digital publications across the globe. While the group also generated social media responses, those efforts failed to gain noticeable traction. In both operations, the malicious actors primarily leveraged AI tools for internal project reporting and to translate or localize state propaganda into multiple regional languages.

For cybersecurity practitioners and media platforms, these findings signal a tactical shift in state-sponsored information warfare. Disinformation actors are moving beyond obvious social media botnets to target editorial pipelines directly. AI models are serving as operational force multipliers, enabling small teams to produce authentic-sounding, translated content that bypasses traditional journalistic vetting.